George Bate posted on Tuesday, July 14, 2020 - 11:17 am
Thank you for your help with this issue. I successfully ran a DSEM. The degrees of freedom are positive, the PSRs are < 1.01, and the results make sense. However, I get this warning message:
WARNING: PROBLEMS OCCURRED IN SEVERAL ITERATIONS IN THE COMPUTATION OF THE STANDARDIZED ESTIMATES FOR SEVERAL CLUSTERS. THIS IS MOST LIKELY DUE TO AR COEFFICIENTS GREATER THAN 1 OR PARAMETERS GIVING NON-STATIONARY MODELS. SUCH POSTERIOR DRAWS ARE REMOVED. THE FOLLOWING CLUSTERS HAD SUCH PROBLEMS:
I'm wondering if the results can be trusted given the warning message, the positive dfs, and PSRs < 1.01. I've included my syntax below. Thank you for all your help.
1. I run two different models and got each time the following warning message: 'WARNING One or more individual-level variables have no variation within a cluster for the following clusters.'
I looked at the data (emotions assessed continuously once a second) there are a lot of variations for each person. How can that be?
2. Furthermore, I try to add two between-level predictors (RSlis RSsha) but then I get the following warning: ' *** FATAL ERROR THE VARIANCE COVARIANCE MATRIX IS NOT SUPPORTED. ONLY FULL VARIANCE COVARIANCE BLOCKS ARE ALLOWED.'
What did I do wrong?
Model: %WITHIN% s1 | emolis ON emolis&1; s2 | emosha ON emosha&1; s12 | emolis ON emosha&1; s21 | emosha ON emolis&1;
%BETWEEN% s12 s21 ON RSlis RSsha; emolis WITH emosha s1 s2 s12 s21 RSlis RSsha; emosha WITH s1 s2 s12 s21 RSlis RSsha; s1 WITH s2 s12 s21 RSlis RSsha; s2 WITH s12 s21 RSlis RSsha; s12 WITH s21; RSlis WITH RSsha;
3. Finally, I run the same analysis with a second data set where more variables are included - and I get different results despite using the identical variables and even after checking it looks all ok. How can that be the case?
1. The data that you looked at isn't the same data that Mplus analyzed. I don't know why there is a difference - possibly look for warning messages in the output file regarding deleted observations due to missing covariates or some other similar thing is happening. You can see the data that Mplus analyzes using savedata: file is 1.dat;
2. To make full blocks use either of these equivalent models
s12 s21 ON emolis emosha s1 s2 RSlis RSsha ; s12 with s21; emolis emosha s1 s2 RSlis RSsha with emolis emosha s1 s2 RSlis RSsha;
3. Bayes estimation uses random draws for drawing numbers from posterior distribution and small changes will occur with such an input change due to changing the order of the random numbers that are used for a particular parameter. It would be similar to changing the seed of the random number generator. Note, however, that this only applies if you have changed the model to include more variables or if the new data set has the same data in a different order. If none of these is true you must look for a data reading problem where the input file doesn't match what is in the data.